Toward phonologically aware language models. Current language models process and “understand” language in a way that is fundamentally different to human cognition that is grounded in spoken dialogue.

Description

Toward phonologically aware language models. Current language models process and “understand” language in a way that is fundamentally different to human cognition that is grounded in spoken dialogue. This project aims to add new knowledge about how phonological features, including intonation, stress, rhythm and systematic sound-meaning mappings contribute to language processing. It will leverage naturalistic speech comprehension and production paradigms with brain imaging to test a new framework that integrates phonological, semantic and syntactic representations in a combined feature space. The outcomes will inform a new generation of “phonologically aware” language models that more closely resemble human cognition, with users benefiting from the models' enhanced processing.. Scheme: Discovery Projects. Field: 5204 - Cognitive and Computational Psychology. Lead: Prof Greig de Zubicaray